Facial Geometrical Features based Pain Assessment using KNN and Regression Tree Classifiers

Abstract

Recent advancements in the field of computer vision and artificial intelligence (AI) have made it possible to detect pain based on facial reactions or changes. Several machine learning and deep learning techniques are employed to identify the pain and code the changes. In this paper, we analyze geometric features obtained from the mouth area using machine learning classifiers such as K Nearest Neighbor (KNN) Classifiers and Regression Trees for pain detection. Biovid Heat Pain database (Part A) was used to develop the proposed system for pain assessment. Pain videos (those without pain and those with high pain) are first divided into individual frames and a set of geometrical and statistical features are extracted. To classify pain, these features are provided to the KNN and RT classifiers. Accordingly, the KNN classifier outperformed the RT classifier at identifying pain from facial features. The most effective features for recognizing pain were geometrical features (area) and a combination of the area- and ratio-based features. The present study will therefore contribute greatly to the design and development of a methodology for the assessment of pain using facial features.